Outline
Layering Principles
- Each layer offers an abstraction of the layers below it.
- Logically, all communication is between peers, protocols
at the same layer on different systems.
- Each layer communicates with and only depends on the layer below
it and on its peers.
OSI Architecture
- Application: user-level program (Web, e-mail, ftp, telnet)
- Presentation: formatting, encryption, data encoding (ASN.1, XDR)
- Session: establishing the endpoints, identification (TCP, UDP)
- Transport: reliability, end-to-end issues (TCP, UDP)
- Network: basic end-to-end connectivity (IP, ATM AAL)
- Data Link: hop-by-hop connectivity, framing (Ethernet, FDDI,
ATM cells)
- Physical: bit transport (fiber, CAT 5, serial)
Internet Architecture
- p. 37 of the textbook
- not strictly layered
- historically came before OSI, or it might have modeled on OSI
- layers:
- device
- IP
- Transport
- application.
- IP "hourglass" allows universal communication
IETF: Internet Engineering Task Force
- "running code" (multiple implementations) is essential
- theoretical definition sometimes takes a bow to the implementation.
- RFCs (request for comments)
- consensus building through technical co-operation
In-class exercises
- where does the Network File System (NFS) fit in on the OSI model? In
the internet model? Does this make sense?
- where does Remote Procedure Call (RPC) fit in on the OSI model? In
the internet model? Does this make sense?
Ethernet
- Aloha net + carrier sense + collision detection
- Carrier Sense Multiple Access/Collision Detection (CSMA/CD)
- variety of implementations, including:
- speeds:
- 3Mb/s: original
- 10Mb/s: most popular today
- 100Mb/s: gaining
- 1Gb/s: new
- addresses:
- 16-bit: original
- 48-bit: current
Traditional Ethernet Organization
- 48-bit addresses are globally unique
- structure of 48-bit addresses not used in forwarding packets
- broadcast domain: every host sees every packet
- single wire carries signal to all hosts
- sender detecting collision sends a jamming signal, does
exponential back-off
Current Ethernet Organization
- hubs: point-to-point wires come in to common bus, where collision detection
occurs (logically shared medium)
- to broadcast, hubs must be interconnected with no loops
- switches: point-to-point wires come in to buffered bus, from which
packets are sent:
- only on wire to destination if known (logically partitioned medium)
- broadcast otherwise (logically shared medium)
- switches might be interconnected with loops (redundant links),
so to broadcast, dynamically compute a spanning tree, a loop-free
subset that reaches every branch
Ethernet Format
- 64-bit preamble: 010101010101....
- 48-bit destination address (all 1's for broadcast)
- 48-bit source address
- 16-bit type field ( 0x800 for IP, 0x806 for ARP)
- 46 to 1500 bytes of data
- 32-bit CRC
- 8-bit postamble
In-class exercises
- What is the length of a bit on a 10Mb/s ethernet? assume
speed of light of 2/3\ c = 200,000\ km/s
- Assume that there are 512 bits in a minimally-sized (46-byte payload)
ethernet frame. What is the length of this frame?
- Compute the maximum length of an ethernet, knowing that a frame
has to be visible at the same time to both the sender and the farthest
receiver.
In-class or at-home exercises
- what is the size (in bytes) of an ethernet header? (header is
source and destination address, and protocol type).
- Ethernet is often used to carry TCP/IP traffic. The TCP/IP traffic
have been carefully designed so all 32-bit fields have 32-bit alignment.
If the Ethernet packet carrying TCP/IP traffic is 32-bit aligned, what is
the alignment of the TCP/IP headers?
- Add an additional 16-bit field to the ethernet header. How many
bytes is the header now? If the ethernet packet is aligned, what is
the alignment of the payload now?
- Suggest some uses for the new 16-bit field you have added to the
ethernet standard.